文献类型：期刊文献

中文题名：基于循环条件生成对抗网络的数据生成方法

英文题名：Data Generation Method Based on Cyclic Conditions to Generative Adversarial Network

作者：孔洁[1]

机构：[1]中国劳动关系学院,北京100048

年份：2021

卷号：46

期号：11

起止页码：134

中文期刊名：火力与指挥控制

外文期刊名：Fire Control & Command Control

收录：CSTPCD、、BDHX、CSCD、BDHX2020、CSCD_E2021_2022

基金：中央高校基本业务费专项基金资助项目(21ZYJG001)。

语种：中文

中文关键词：生成对抗网络;长短期记忆网络;循环神经网络;数据生成;军事数据

外文关键词：generative adversarial network;long and short term memory network;recurrent neural network;data generation;military data

中文摘要：针对军事领域数据采集困难以及数据生成人工依赖性强、成本高、效率低等问题,在生成对抗网络(GAN)框架的基础上,构建长短期循环神经网络(LSTM-RNN)代替生成对抗网络中的生成器和鉴别器,以最大平均差异和最大似然估计作为指标构建数据生成评估模型,提出一种可生成数据序列的循环条件生成对抗网络(RCGAN)。该方法完全依靠数据驱动,无需经过精心设计的建模过程,便可生成与真实数据相一致的数据。通过基于实际数据的仿真实验,验证了该方法在数据相似度、估计误差、抗干扰性以及泛化性方面的优势。

外文摘要：Aiming at the difficulties of data acquisition in military field and such problems as strong dependence on human,high cost and low efficiency,etc.of data generation,based on generative adversarial network(GAN)framework,the long and short term cyclic neural network(LSTM-RNN)is constructed to replace the generator and discriminator in the generative adversarial network.The maximum average difference and maximum likelihood estimation are used as indicators to construct the data generation evaluation model,and a cyclic condition to generative adversarial network(RCGAN)that can generate data sequences is proposed.This method is completely dependent on data-driven and can generate data consistent with the real data without a carefully designed modeling process.The simulation experiments based on actual data verify the advantages of this method in data similarity,estimation error,anti-interference and generalization aspects.